[CI] Mark 1-GPU L4 test steps with device: l4 for EKS migration (#54549)
Signed-off-by: khluu <khluu000@gmail.com>
Kevin H. Luu committed
Aug 31, 2026 at 14:56 UTC
24d42f3553a35aa53006a3ee8d04cc9191cb7be9
5 files changed
+15
.buildkite/test_areas/distributed.yaml
+1
@@ -20,6 +20,7 @@ steps:
20
21
- label: ":nvidia: (L4) Sharded RDT Weight Transfer"
22
key: sharded-rdt-weight-transfer
23
+ device: l4
24
timeout_in_minutes: 20
25
working_dir: "/vllm-workspace/tests"
26
num_devices: 1
.buildkite/test_areas/engine.yaml
+2
@@ -38,6 +38,8 @@ steps:
38
39
- label: ":nvidia: (L4) V1 Engine"
40
key: engine-1-gpu
41
+ device: l4
42
+ num_devices: 1
43
timeout_in_minutes: 45
44
source_file_dependencies:
45
- vllm/v1/engine/
.buildkite/test_areas/kernels.yaml
+6
@@ -104,6 +104,8 @@ steps:
104
105
- label: ":nvidia: (L4) Attention Kernels Shard %N"
106
key: kernels-attention-test
107
+ device: l4
108
+ num_devices: 1
109
timeout_in_minutes: 65
110
source_file_dependencies:
111
- csrc/attention/
@@ -165,6 +167,8 @@ steps:
167
168
- label: ":nvidia: (L4) Quantization Kernels Shard %N"
169
key: kernels-quantization-test
170
+ device: l4
171
+ num_devices: 1
172
timeout_in_minutes: 60
173
source_file_dependencies:
174
- csrc/quantization/
@@ -194,6 +198,8 @@ steps:
198
199
- label: ":nvidia: (L4) MoE Kernels Shard %N"
200
key: kernels-moe-test
201
+ device: l4
202
+ num_devices: 1
203
timeout_in_minutes: 50
204
source_file_dependencies:
205
- csrc/quantization/cutlass_w8a8/moe/
.buildkite/test_areas/models_language.yaml
+2
@@ -92,6 +92,8 @@ steps:
92
# correctness test on L4 until its H200 output matches the Transformers reference.
93
- label: ":nvidia: (L4) Granite Language Model Compatibility"
94
key: language-models-tests-granite-l4-compatibility
95
+ device: l4
96
+ num_devices: 1
97
timeout_in_minutes: 65
98
source_file_dependencies:
99
- vllm/
.buildkite/test_areas/pytorch.yaml
+4
@@ -90,6 +90,8 @@ steps:
90
91
- label: ":nvidia: (L4) PyTorch Compilation Passes"
92
key: pytorch-compilation-passes-unit-tests
93
+ device: l4
94
+ num_devices: 1
95
timeout_in_minutes: 45
96
source_file_dependencies:
97
- vllm/__init__.py
@@ -164,6 +166,8 @@ steps:
166
# those SM90 cases skip while the architecture-compatible cases still run.
167
- label: ":nvidia: (L4) PyTorch Fullgraph CUDAGraph Compatibility"
168
key: pytorch-fullgraph-cudagraph-l4-compatibility
169
+ device: l4
170
+ num_devices: 1
171
timeout_in_minutes: 60
172
source_file_dependencies:
173
- vllm/__init__.py